{"id":"W4385573178","doi":"10.18653/v1/2022.blackboxnlp-1.10","title":"How (Un)Faithful is Attention?","year":2022,"lang":"en","type":"article","venue":"","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Computer science; Metric (unit); Artificial intelligence; Binary number; Encoder; Simple (philosophy); Measure (data warehouse); Machine learning; Data mining; Mathematics; Arithmetic","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002973709,0.0005349735,0.0005267014,0.001146786,0.000592267,0.002557204,0.00121714,0.001660932,0.002382728],"category_scores_gemma":[0.03674247,0.0004816398,0.0003786513,0.0006734963,0.002757039,0.005081476,0.001455286,0.002188509,0.0003841699],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000961087,"about_ca_system_score_gemma":0.0005461393,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002958308,"about_ca_topic_score_gemma":0.002918592,"domain_scores_codex":[0.998329,0.0005894396,0.00009727874,0.0005464319,0.000289257,0.0001486272],"domain_scores_gemma":[0.9865414,0.00769343,0.001354663,0.00288611,0.001058527,0.0004658364],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001014759,0.0002211272,0.06026852,0.001009513,0.0009310583,0.0007317627,0.004633802,0.1015226,0.04786456,0.2585657,0.006743632,0.5164929],"study_design_scores_gemma":[0.00005642344,0.0001889661,0.0341399,0.0002357447,0.0002432817,0.0006808534,0.001232538,0.2283492,0.01628521,0.7089739,0.00951414,0.00009983954],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4263631,0.003401199,0.5457842,0.006996384,0.0002741569,0.00009427164,0.0007318992,0.001254339,0.01510057],"genre_scores_gemma":[0.9871573,0.0002975338,0.01120744,0.0003307318,0.000058921,0.00002271174,0.0001305239,0.00009700581,0.0006978608],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.002973709,"threshold_uncertainty_score":0.01572663,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02181393702308317,"score_gpt":0.2358932240045686,"score_spread":0.2140792869814855,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}